Wave propagation and imaging in random waveguides - Liliana Borcea Panoramas et Synthèses

Page created by Alexander Logan
 
CONTINUE READING
Wave propagation and imaging in random waveguides - Liliana Borcea Panoramas et Synthèses
Wave propagation and imaging
    in random waveguides

                     Liliana Borcea
                            edited by
                   H. Ammari, J. Garnier

                  Panoramas et Synthèses

Numéro 44

SOCIÉTÉ MATHÉMATIQUE DE FRANCE
Publié avec le concours du Centre national de la recherche scientifique
Wave propagation and imaging in random waveguides - Liliana Borcea Panoramas et Synthèses
Panoramas & Synthèses
44, 2014, p. 1–61

                 WAVE PROPAGATION AND IMAGING
                    IN RANDOM WAVEGUIDES

                                                 by

                                        Liliana Borcea

       Abstract. – We present a rigorous asymptotic analysis of wave propagation in waveg-
       uides with random boundaries and with random fluctuations of the wave speed. We
       consider scalar waves in both two and three dimensions. The asymptotic analysis
       is with respect to the small magnitude of the random fluctuations, as they occur
       naturally in applications like underwater acoustics. Cumulative scattering effects of
       such fluctuations become significant at long distances of propagation. They couple
       the waveguide modes and cause loss of coherence of the wave field. We quantify ex-
       plicitly mode coupling and loss of coherence via three important length scales: the
       scattering mean free path, the transport mean free path, and the equipartition dis-
       tance. Knowledge of these scales is important for improving imaging methods and
       understanding their limitations. We consider a source imaging problem in a random
       waveguide. The data are the signals recorded by a receiver array far from the source.
       We study coherent array imaging and contrast it with the time-reversal process. We
       show that scattering by the random inhomogeneities is beneficial in general for the
       time reversal process, but it always impedes imaging. The resolution and robustness
       of the images deteriorate as the waves travel farther in the waveguide from the source
       to the array. At distances that exceed the scattering mean free path of all the modes,
       the wave field becomes incoherent and imaging can only be done with parametric
       estimation methods based on models of transport of energy. We analyze in detail the
       degradation of images in random waveguides, and illustrate the results with numerical
       simulations.

       Résumé. – Nous présentons une analyse asymptotique rigoureuse de la propagation
       des ondes dans des guides d’ondes avec des bords aléatoires et avec des fluctuations
       aléatoires de la vitesse de propagation. Nous considérons des ondes scalaires en deux
       et trois dimensions. L’analyse est asymptotique par rapport à la faible amplitude
       des fluctuations aléatoires, comme il se produit naturellement dans des applications
       comme l’acoustique sous-marine. Les effets cumulatifs de diffusion dues à ces fluc-
       tuations deviennent importants sur de longues distances de propagation. Ils se ma-
       nifestent par un couplage des modes du guide et une perte de cohérence du champ
       d’onde. Nous quantifions explicitement le couplage de modes et de la perte de cohé-
       rence par l’intermédiaire de trois échelles de longueur importantes : le libre parcours

2010 Mathematics Subject Classification. – 76B15, 35R60, 60F05, 35R30.
Key words and phrases. – Imaging, wave propagation, random media, asymptotic analysis.

                                                           © Panoramas et Synthèses 44, SMF 2014
2                                      LILIANA BORCEA

      moyen de diffusion, le libre parcours moyen de transport et la distance d’équiparti-
      tion. La connaissance de ces échelles de longueur est importante pour améliorer les
      méthodes d’imagerie et pour comprendre leurs limites. Nous étudions un problème
      d’imagerie de source dans un guide d’ondes aléatoire. Les données sont les signaux
      enregistrés par un réseau de récepteurs situés loin de la source. Nous considérons un
      procédé d’imagerie cohérente par migration et nous la comparons avec le processus
      de retournement temporel. Nous montrons que la diffusion par les inhomogénéités
      aléatoires est bénéfique en général pour le processus de retournement temporel, mais
      il détériore toujours l’imagerie. La résolution et la robustesse des images se dégradent
      au fur et à mesure que la distance entre la source et le réseau de récepteurs augmente.
      A des distances supérieures au libre parcours moyen de diffusion de l’ensemble des
      modes, le champ d’ondes est incohérent et l’imagerie ne peut être effectuée par des
      méthodes d’estimation paramétrique basées sur des modèles de transport d’énergie.
      Nous analysons en détail la dégradation des images dans des guides d’ondes aléatoires,
      et illustrons les résultats par des simulations numériques.

                                         1. Introduction
   Array imaging is an important technology with a wide range of applications in
underwater acoustics, seismology, non-destructive evaluation of materials, medical ul-
trasound, and elsewhere. It is concerned with locating remote, compactly supported
sources and/or scatterers from measurements at arrays of sensors. The sensors are
devices that transform one form of energy into another. Depending on the application
they may be antennas that convert electromagnetic waves to/from electric signals,
hydrophones that convert changes in water pressure to electrical signals, ultrasonic
transducers that transmit and receive ultrasound waves, and so on. When the sensors
are located sufficiently close together, they behave as a collective entity, called the
array. The sensors in passive arrays are receivers of the waves generated by unknown
remote sources. Active arrays have sensors that play the dual role of sources and re-
ceivers. The sources emit waves that propagate through the medium and are scattered
back to the array, where they are captured by the receivers.
   The recordings at the receivers are called the array data. The coherent imaging
process seeks to transform the data into an imaging function that peaks in the sup-
port of the unknown sources or scatterers. The key data processing step in the image
formation is the synchronization of the received signals using a mathematical model
of wave back-propagation from the array to a search (imaging) location ~rs . The ex-
pectation is that when ~rs lies in the support of the unknown sources or scatterers,
the recordings are synchronized and add up over the sensors to give a peak value
of the imaging function. Indeed, this happens if we have an accurate model of wave
propagation between the array and the imaging region.
   Image formation is somewhat similar to the time reversal process. Time reversal
is an experiment that uses an active array to first receive the waves from a remote
source, and then re-emit the time-reversed recordings back in the medium. The wave

PANORAMAS & SYNTHÈSES 44
WAVE PROPAGATION AND IMAGING IN RANDOM WAVEGUIDES                            3

equation is time reversible if the medium is non-dissipative, so the waves are expected
to propagate back to the source and focus near it. The focusing resolution in the range
direction is mostly affected by the bandwidth of the emitted signals. The focusing in
cross-range (i.e., in the plane orthogonal to the range) depends on many factors, such
as how large the array is, how far the source is, and how much scattering occurs
between the source and receiver. The focusing is never perfect, because the array
cannot capture the whole wave field emitted from the source. Thus, the time reversal
process is not exactly like solving the wave equation backward in time to recover the
initial source. The expectation is that the larger the array is, the better the focus,
because more of the waves are captured and turned back. But no matter how large
the array is, even if it surrounds the source, the evanescent waves cannot reach it,
and the focusing resolution is limited by diffraction. For example, the resolution of
refocusing of time harmonic waves cannot exceed the Abbe diffraction limit of λ/2,
where λ is the wavelength [12]. In most applications the array is supported in a set
of small diameter (aperture) | A | with respect to the distance L of propagation, and
the expected resolution of refocusing in cross-range is given by the Rayleigh formula
λL/| A | if there is no scattering in the medium. The interesting property of time
reversal, known as super-resolution, is that scattering may improve the cross-range
focusing. This is demonstrated and explained in [15, 16, 26], and it is analyzed
theoretically and numerically in [5, 4, 30]. The latter studies are based on random
models of the scattering medium, and introduce the important concept of statistical
stability. Stability means that the focusing is robust, it is independent of the realization
of the random medium, it is observable and reproducible. Stability is guaranteed in
general only for sources emitting broad-band signals, as pointed out in [5], although
there are regimes where it holds for narrowband signals as long as the sources and
arrays have sufficiently large support [30].
   The super-resolution and robustness of time reversal are due entirely to the waves
propagating back in exactly the same medium they came from. The observer has
access to the vicinity of the source and can see the focus there, whereas in imaging
the access is limited to the array. Moreover, the medium is not known in detail in many
imaging applications. For example, in underwater acoustics, the smooth part of the
wave speed is known, but there are small scale fluctuations due to internal waves [17]
that cannot be known, and are not even of interest in imaging. It is the uncertainty
about such small scale inhomogeneities that we model with random spatial processes,
and thus speak of imaging in random media.
   The array data processing for image formation may appear to mimic the back-
propagation of the waves in the time reversal process, but there is a fundamental
difference. While in time reversal the waves go back in the same medium, the back-
propagation in imaging is done mathematically, using a fictitious medium model that
incorporates the large scale features of the wave speed, but not its fluctuations, which
are unknown. This difference is essential in regimes with strong cumulative scattering
of the waves by the random inhomogeneities. The longer the waves travel in the ran-
dom medium, the stronger the scattering, which leads to loss of coherence of the wave

                                                           SOCIÉTÉ MATHÉMATIQUE DE FRANCE
4                                   LILIANA BORCEA

field. The coherent waves are useful in imaging because we can relate the locations
of the sought-after sources or scatterers to their arrival time. Multiple scattering by
the inhomogeneities transfers energy to the incoherent part of the waves, the random
fluctuations which are unwanted reverberations. We may think of the reverberations
as noise, but we should remember that they are not ordinary additive, uncorrelated
noise. They have complicated statistical structure, with correlations across the array
and in the bandwidth, and are difficult to mitigate. Our goal is to quantify the de-
terioration of resolution and robustness of images in random media, and to propose
efficient methods for mitigating cumulative scattering effects.
    The data processing in image formation must take into account the presence of
scattering boundaries, because they create multiple traveling paths of the waves re-
ceived at the array. This happens in waveguides, where boundaries trap the waves
and guide the energy in a preferred direction, the waveguide axis, along which the
medium is unbounded. To analyze the wave field in waveguides, we may decompose it
mathematically in an infinite, countable set of monochromatic waves called waveguide
modes. In ideal waveguides, with flat boundaries and wave-speed that is constant or
varies smoothly in the waveguide cross-section, the modes are uncoupled. Finitely
many of them propagate along the waveguide axis at different speeds, and we may
associate them to plane waves with different angles of incidence at the boundary. The
slower modes correspond to nearly normal incidence. They bounce off the bound-
aries many times, thus taking a long path from the source to the array. The faster
modes correspond to small grazing angles at the boundary, and shorter paths to the
array. The remaining infinitely many modes are evanescent, with amplitudes decaying
exponentially with the distance along the waveguide axis.
    We are concerned with waveguides that are randomly perturbed versions of the
ideal ones. They may have boundaries with small random fluctuations, and/or internal
inhomogeneities that cause small fluctuations of the wave speed. We present a rigorous
asymptotic theory of wave propagation in such waveguides, where the asymptotics is
in the small parameter that measures the magnitude of the fluctuations. We show that
when the waves travel sufficiently far from the source, the small fluctuations cause
significant cumulative scattering. It amounts to coupling of the waveguide modes,
and gradual loss of coherence of their amplitudes. We give a detailed analysis of
mode coupling and quantify the net scattering via three important length scales: the
scattering mean free path, the transport mean free path and the equipartition distance.
The mode dependent scattering mean free path is the distance over which the mode
loses its coherence, meaning that its random fluctuations dominate its statistical mean.
The transport mean free path is classically defined as the distance beyond which the
waves forget their initial direction [31]. Because the modes are associated to directions
of incidence at the boundary, we define the mode dependent transport mean free path
as the length scale over which the mode exchanges energy with the other modes. The
equipartition distance is the longest of the three length scales, and it gives the distance
over which scattering distributes the energy uniformly among the modes, independent
of the initial source excitation.

PANORAMAS & SYNTHÈSES 44
You can also read